Blind Signal Recognition Algorithm Based On Statistical Pattern Recognition

As a result of the rapid devolvement of wireless communication, a growing number of wireless communication protocols and signal modulation format utilized in practice, which puts higher requirement on radio monitoring. As one of the core tasks of radio monitoring, automatic modulation classification (AMC) becomes an important research area. Current likelihood base AMC algorithms, feature based AMC algorithms and heuristic AMC algorithms assume there is only one, thus they cannot handle multiple overlapped signal sources. To tackle this problem, the paper proposes a separation and classification algorithm (SCA) to conduct AMC for multiple overlapped signal sources. AMC consists of a separation stage and a classification stage. In the separation stage, the channel matrixes from each source to multiple receiving antennas are estimated using Fast ICA (Fast Independent Component Analysis) by maximizing entropy. Then the individual sources are reconstructed from received signal samples using the estimated channel matrixes. In the classification stage, multiple cumulants are utilized jointly to boost performance over traditional single cumulant AMC. The theoretical cumulant values of candidate modulation formats are computed beforehand, and multiple cumulants with different orders are organized into a feature vector for each one signal. The pattern of the target signal is identified as the candidate signal with a feature vector most similar to that of the target signal. Moreover, since phase offset and frequency offset are common in practice, an analysis of the influence of frequency offset and phase offset on the cumulant based AMC is provided, which is followed by some phase offset and frequency offset estimation algorithms. So as to confirm the performance of the algorithms all above, extensive simulation is also conducted.